Luo-Zhengding / Unsupervised-GFANC

Unsupervised learning based end-to-end delayless generative fixed-filter active noise control

Unsupervised-GFANC: Unsupervised Learning Based End-to-End Delayless Generative Fixed-Filter Active Noise Control

This repository contains the code for the paper "Unsupervised Learning Based End-to-End Delayless Generative Fixed-Filter Active Noise Control," accepted by the 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2024). The paper is available on ArXiv and IEEE Xplore.

Highlights

1. Simplified Training Process: The 1D CNN in our unsupervised-GFANC method does not require initial training using labelled noise data, simplifying the training process and enhancing practicality.
2. End-to-End Differentiable ANC System: Integrating the co-processor and real-time controller into the ANC system allows for using the accumulated squared error signal as the loss for training the 1D CNN.
3. Improved Noise Reduction Performance: The unsupervised-GFANC method exhibits better noise reduction performance than the supervised-GFANC method by avoiding labelling errors.

Usage

1. Pre-trained Model: If you don't want to retrain the 1D CNN (M5_Network.py), you can use the trained model available in models/1DCNN_SyntheticDataset_UnsupervisedLearning.pth. Simply run the Noise_Cancellation_RealNoise_RealPath.ipynb notebook to get the noise reduction results.
2. Training Dataset: The 1D CNN is trained using a synthetic noise dataset with label files Soft_Index.csv. The entire dataset is available here.
3. Applying to New Acoustic Paths: We have provided the sub control filters on synthetic acoustic paths and our measured acoustic paths. If you want to use the Unsupervised-GFANC method on new acoustic paths, just obtain the corresponding pre-trained broadband control filter and decompose it into sub control filters. The trained 1D CNN in Unsupervised-GFANC can remain unchanged. For more details, please refer to the paper.

Related Works

• Deep Generative Fixed-Filter Active Noise Control
• Delayless Generative Fixed-filter Active Noise Control based on Deep Learning and Bayesian Filter
• GFANC-Kalman: Generative Fixed-Filter Active Noise Control with CNN-Kalman Filtering
• A hybrid sfanc-fxnlms algorithm for active noise control based on deep learning
• Performance Evaluation of Selective Fixed-filter Active Noise Control based on Different Convolutional Neural Networks

If you are interested in our works, please consider citing our papers. Thanks for your interest! Have a great day!